Surface velocities in the hinterland of the Neumayer III station (Antarctica)

(1)

Surface velocities in the hinterland of the Neumayer III station (Antarctica)

derived from SAR-Interferometry

Niklas Neckel

(2)

(3)

Universität Heidelberg

Fakultät für Chemie und Geowissenschaften

Geographisches Institut

Diplomarbeit/Diploma Thesis

Surface velocities in the hinterland of the Neumayer III station (Antarctica)

derived from SAR-Interferometry

Author:

Niklas Neckel Herkenkrug 20 22359 Hamburg

Referee:

Prof. Dr. Lucas Menzel PD Dr. Olaf Eisen

Supervisor:

Dipl.-Phys. Reinhard Drews Dr. Wolfgang Rack

December 21, 2010

(4)

(5)

Ich versichere, dass ich die beiliegende Diplomarbeit ohne Hilfe Dritter und ohne Benutzung anderer als der angegebenen Quellen und Hilfsmittel angefertigt und die

den benutzten Quellen wörtlich oder inhaltlich entnommenen Stellen als solche kenntlich gemacht habe. Diese Arbeit hat in gleicher oder ähnlicher Form noch

keiner Prüfungsbehörde vorgelegen.

Heidelberg, den 21.12.2010

(6)

(7)

Abstract

Surface velocities of polar ice are an important input parameter for mass ux calcu- lations and ice-sheet modelling. As on-site measurements in remote areas are sparse, satellite-based measurements have to be used to obtain area-wide surface velocities.

Synthetic Aperture Radar (SAR) data from various sensors are routinely employed for this purpose. Depending on the availability of adequate SAR image pairs, the surface velocity can be derived by SAR interferometry. The accuracy of the applied interferometric method heavily depends on external input parameters (e.g. elevation model) and the processing history.

The present thesis focuses on the hinterland of the German overwintering station Neumayer III (Antarctica) and complements pre-site surveys for a future deep drill ice core site. The dependency of the interferometric approach on external elevation models is tested by comparing surface velocities based on Antarctic-wide elevation models (from satellite altimetry) with surface velocities based on local elevation models (from SAR interferometry). The accuracy of the generated surface velocity elds is evaluated by comparing the data with on-site GPS measurements.

A map of surface velocities in the hinterland of the German overwintering station Neumayer III and a precise estimate of the grounding zone location are presented as geophysical results. The derived surface velocities cover an area of ∼17.000 km² of oating and grounded ice and are based on 16 SAR scenes from the European Remote-Sensing Satellites (ERS) 1/2 acquired between 1994-1996. The derived surface velocities vary between 0 m/d 0.5 m/d with a locally varying error between 0.002 m/d 0.08 m/d.

(8)

(9)

Zusammenfassung

Oberächengeschwindigkeiten polarer Eismassen sind ein wichtiger Parameter für die Modellierung von Eisschilden und deren Massenbilanzierung. Die allgemeine Unzu- gänglichkeit des antarktischen Kontinents erschwert groÿächige Messungen vor Ort, weshalb vermehrt Satelliten gestützte Methoden verwendet werden. Unter anderem werden Radar-Satellitensysteme mit Synthetischer Apertur (engl. Synthetic Aperture Radar oder SAR) verwendet. Wenn SAR Bilder mit hinreichender Qualität verfügbar sind, können Oberächengeschwindigkeiten mittels Radarinterferometrie bestimmt werden. Die Genauigkeit der hier angewendeten Methode hängt stark von externen Parametern (z.B. digitalen Geländemodellen) und den einzelnen Prozessierungsschrit- ten ab.

Der regionale Fokus dieser Diplomarbeit liegt südlich der deutschen Überwinterungssta- tion Neumayer III (Antarktis) und ergänzt die Vorerkundungen für eine neue Eiskern Tiefenbohrung. In der vorliegenden Arbeit werden Geschwindigkeitsfelder basierend auf unterschiedlichen Geländemodellen (Antarktis weite Geländemodelle, die auf Satel- liten Altimetrie Messungen beruhen, sowie regionale Geländemodelle, die auf Radar- interferometrie basieren) generiert und verglichen. Um die Genauigkeit der abgeleit- eten Flieÿgeschwindigkeiten abschätzen zu können, werden diese, soweit möglich, mit lokalen GPS Geschwindigkeitsmessungen verglichen.

Als geophysikalisches Ergebnis werden sowohl eine Karte von Oberächengeschwindig- keiten des Untersuchungsgebiet als auch eine genaue Kartierung der Aufsetzzone präsentiert. Die Inland- und Schelfeis Geschwindigkeiten konnten für eine Fläche von ca. 17.000 km² berechnet werden. Hierfür wurden 16 SAR Szenen verwendet die zwischen 1994-1996 von ERS-1 und ERS-2 (ERS - engl. European Remote-Sensing Satellite) aufgenommen wurden. Die berechneten Oberächengeschwindigkeiten re- ichen von 0 m/t 0.5 m/t und weisen einen örtlichen Fehler zwischen 0.002 m/t 0.08 m/t auf.

(10)

(11)

List of Figures

1.1. Region of interest . . . 4

1.2. Depiction of the Antarctic ice system (modied after Bell (2009)) . . 6

1.3. Illustration of the grounding zone (after Fricker and Padman (2006)) 7 2.1. Relation between phase, amplitude and wavelength . . . 11

2.2. The principle of a SAR system . . . 12

2.3. Setup for interferometric imaging . . . 14

2.4. Sensitivity of ERS to vertical and horizontal motion (after Meyer (2004)) 17 3.1. ERS satellite tracks used for InSAR processing . . . 20

3.2. Elevation dierences along airborne laser altimeter proles . . . 27

3.3. Location of proles . . . 28

3.4. Laser altimetry prole . . . 29

3.5. GPS prole . . . 29

4.1. Work ow in the production of a three-dimensional velocity eld . . . 32

4.2. Interferogram before and after at-earth removal . . . 35

4.3. Interferogram. Fringes caused by topography, surface displacement and tidal movement . . . 36

4.4. Interferogram before and after subtracting topography . . . 38

4.5. Interferogram before and after phase unwrapping . . . 39

4.6. Three-dimensional illustration of the relation between GPS-derived velocity and the velocity along the satellite's LOS . . . 40

4.7. Unwrapped `motion-only' interferogram . . . 42

4.8. One-dimensional ow eld of an ascending and a descending satellite path . . . 43

4.9. Three-dimensional velocity eld . . . 44

5.1. Mosaic of dierences in overlapping areas . . . 48

5.2. Dierences between two three-dimensional velocity elds in overlapping area . . . 50

5.3. Dierences between aspect angle and ow direction . . . 52

5.4. Comparison between calculated ow and hill slope . . . 54

5.5. Comparison between surface velocities based on dierent DEMs . . . 55

(14)

List of Figures

5.6. Location of GCPs used for adjustment and evaluation of the calculated

velocity eld . . . 57

5.7. Location of proles in the region of the main ice ow . . . 59

5.8. Prole 2 in the region of the main ice ow . . . 59

6.1. Grounding line detection from dierent sources . . . 62

6.2. Ice ow in the Neumayer III hinterland . . . 64

A.1. Basic structure of the Python script 3D_DISP.py . . . 67

A.2. Proles along and across an ice stream . . . 69

A.3. Prole along an ice stream . . . 70

A.4. Prole across an ice stream . . . 70

(15)

List of Tables

1.1. Cryospheric components (modied after Lemke et al. (2007)) . . . 1

3.1. ERS satellite tracks used for InSAR processing . . . 21

3.2. GPS-derived ow vectors (modied after Riedel (2002)) . . . 24

3.3. Available DEMs . . . 25

5.1. Comparision of calculated ow velocities with GPS measurements . . 58

6.1. Three-dimensional velocity elds used for the nal mosaic . . . 63

(16)

(17)

List of Abbreviations

ALS . . . Airborne Laser Scanner AMI . . . Active Microwave Instrument

ASAID . . . Antarctic Surface Accumulation and Ice Discharge

ASTER . . . Advanced Spaceborne Thermal Emission and Reection Radiome- ter

AWI . . . Alfred-Wegener-Institute for Polar and Marine Research DEM . . . Digital Elevation Model

DEOS . . . Delft Institute for Earth-oriented Space Research DML . . . Dronning Maud Land

ERS . . . European Remote-Sensing Satellite ESA . . . European Space Agency

GCP . . . Ground Control Point

GLAS . . . Geoscience Laser Altimeter System GPR . . . Ground Penetrating Radar

GPS . . . Global Positioning System

GRACE . . . Gravity Recovery and Climate Experiment ICESat . . . Ice, Cloud and Land Elevation Satellite InSAR . . . Interferometric SAR

IPCC . . . Intergovernmental Panel on Climate Change LIMA . . . Landsat Image Mosaic of Antarctica

(18)

List of Abbreviations

LIMPICS . . . Linking micro-physical properties to macro features in ice sheets with geophysical techniques

LOS . . . Line Of Sight

METI . . . Ministry of Economy, Trade and Industry of Japan MOA . . . MODIS Mosaic of Antarctica

MODIS . . . Moderate Resolution Imaging Spectroradiometer NASA . . . National Aeronautics and Space Administration NH . . . Northern Hemisphere

NSIDC . . . National Snow and Ice Data Center RAMP . . . Radarsat Antarctic Mapping Project RAR . . . Real Aperture Radar

RES . . . Radio Echo Sounding RMSE . . . Root Mean Square Error SLC . . . Single Look Complex SLE . . . Sea Level Equivalent SNR . . . Signal to Noise Ratio

SRTM . . . Space Shuttle Radar Topography Mission TWT . . . Two Way Travel Time

USGS . . . U.S. Geological Survey

(19)

1. Introduction

The world's climate system consists of the compartments atmosphere, cryosphere, hydrosphere, lithosphere and biosphere (Dyck and Peschke 1995, p. 21). These compartments interact in a complex and sensitive way, causing the system's nonlinearity.

Changes between the dierent compartments in this system, fostered by human inter- ference, have unforeseeable eects for the world's climate. Among the most discussed developments is the rise of the global sea level, which is of great importance to both nature and society. In the last Intergovernmental Panel on Climate Change (IPCC) report, the average sea level rise for the 20th century is denoted with 1.7±0.5 mm/a (Bindo et al. 2007, p. 387). This development is attributed to increasing temper- atures, causing a thermal expansion of water on the one hand and higher melting and evaporation rates on the other. As a result, there is an increasing inow of fresh water into the oceans from the melting of the land ice masses of the polar regions and the higher mountains. The potential impact of the cryospheric components on the global sea level rise is listed in Table 1.1.

Table 1.1.: The area, volume and sea level equivalent (SLE) for the cryospheric components are given below. The annual minimum and maximum for snow, sea ice and seasonally frozen ground is shown, as well as the annual mean for the other components (partially only for the Northern Hemisphere (NH)). The sea ice area is represented by the region within the sea ice edge. Modied after Lemke et al. (2007, p. 342).

Cryospheric Component Area (10⁶km²) Ice Volume (10⁶km³) Potential Sea Level Rise (SLE) (m)¹

Snow on land (NH) 1.945.2 0.00050.005 0.0010.01

Sea ice 1927 0.0190.025 0

Glaciers and ice caps² 0.54 0.13 0.37

Ice shelves³ 1.5 0.7 0

Greenland ice sheet⁴ 1.7 2.9 7.3

Antarctic ice sheet³ 12.3 24.7 56.6

Seasonally frozen ground (NH)⁵ 5.948.1 0.0060.065 0

Permafrost (NH)⁶ 22.8 0.0110.037 0.030.10

1Assuming an oceanic area of 3.62 x 10⁸km², an ice density of 917 kg m⁻³, a seawater density of 1.028 kg m⁻³, and seawater replacing grounded ice below sea level.

2Dyurgerov and Meier (2005); glaciers and ice caps surrounding Greenland and Antarctica are excluded.

3Lythe and Vaughan (2001).

4Bamber et al. (2001).

5Zhang et al. (2003).

6Zhang et al. (1999), excluding permafrost under ocean, ice sheets and glaciers.

(20)

While it is evident that the melting of the ice sheet covering Antarctica would have the largest impact, it is also the most unlikely scenario. The melting of mountain glaciers and the Greenland ice sheet, however, is a more realistic threat (Rahmstorf and Schellnhuber 2007, p. 61). The Antarctic ice sheet stores more than 80% of the world's fresh water resources (Roland 2009, p. 241). As shown in Table 1.1, the unlikely collapse of the entire Antarctic ice sheet would result in a sea level rise of 56.6 m. However, also small changes in the overall Antarctic ice mass balance can strongly inuence the global system.

In addition to changes in the sea level, melting of the Antarctic ice is expected to have serious consequences on the world's radiation budget. This is due to the loss of the high albedo of the Antarctic ice (Lemke et al. 2007, p. 341). Also, it is assumed that the thermohaline circulation would be disturbed by the melting of Antarctic ice and the resulting inux of freshwater into the ocean (Massom and Lubin 2006, p. 3).

The mass balance of an ice sheet is quantied in terms of accumulation¹ versus ablation² (Massom and Lubin 2006, p. 11). An imbalance between inputs (e.g.

snowfall) and loss (e.g. calving events, sublimation/evaporation, melt runo) results in a change of mass balance. For mountain glaciers and the Greenland ice sheet melting and evaporation are the main ablation processes (Bentley and Thomas 2007, p. 102). In contrast, Antarctica loses ice mainly due to calving events at the coasts and along the shelf ice edge (Wilhelm 1975, p. 215). Whether the mass balance of the Antarctic ice sheet is positive or negative is not certain at the moment and is strongly dependent on the method applied. Chen et al. (2009, p. 859) estimate a total mass loss of 190 ± 77 Gt/a based on the Gravity Recovery and Climate Experiment (GRACE). Estimates from GRACE data are assumed to be the most accurate at the moment. The acquisition of the individual parameters inuencing the mass balance of the large ice sheets remains a dicult task. This applies in particular for the acquisition of ground truth data (Eisen et al. 2008).

The dynamic response time of the polar ice sheets to climate change is much longer than for small mountain glaciers (Massom and Lubin 2006, p. 6). As a consequence, the response of the world's ice sheets to a global climate change remains the largest unknown parameter for the prediction of the future sea level rise (Massom and Lubin 2006, p. 10). Therefore, it is essential to monitor the mass balance of the Antarctic ice sheet more properly. Several programs pursue this target, including the European ice2sea (see, for example, Vaughan (2009)). A large potential for measurements on a wider scale is given by remote sensing techniques from airplanes and satellites.

Since the ice loss of the Antarctic ice sheet is a highly dynamic process, knowledge of the ice ow is essential for an understanding of the current state and for predicting future developments. As pointed out in Table 1.1, mass changes of ice shelves and

1Accumulation: Positive mass budget, gained primary through snowfall.

2Ablation: Negative mass budget. Mass is lost by a number of mechanisms, which strongly depend on the local environment (e.g. supra and subglacial melting, wind erosion, iceberg calving etc.).

(21)

1. Introduction sea ice have no direct eect on the sea level, as they are not grounded. However, a removal of the ice shelves could speed up the displacement of grounded ice into the ocean, since they act like a stopper for the grounded ice sheet (Massom and Lubin 2006, p. 9). The potential force of this phenomena became apparent when the northern Larsen ice shelf on the Antarctic Peninsula collapsed in 1995 (Rott et al.

2002, 2010).

1.1. Goals of this study

The goals of this thesis are:

• to understand the diculties associated with the interferometric processing used to derive surface velocities and to identify critical steps in this process.

• to automate the processing chain in order to nd many SAR image pairs and to exchange processing parameters more easily.

• to analyze the dependency of the interferometric approach on external elevation models.

• to derive an area-wide velocity eld with error estimates in the region of interest.

• to derive an estimate of the grounding zone location in the region of interst.

1.2. Region of interest

The region of interest (Figure 1.1) is situated in Dronning Maud Land (DML), Antarc- tica. Surface velocities could be derived for the region between 11^◦ W - 6.6^◦ W and 70.7^◦ S - 72.8^◦ S. Large parts of the survey area were rst mapped during the Nor- wegianBritishSwedish Antarctic Expedition (1949-1952) (Riedel 2002, p. 63). The region around the Ekströmisen has a long tradition in German polar research as the overwintering stations Neumayer I-III were located there. The Ekströmisen is named after the Swedish engineer Bertil Ekström, who drowned after he dropped down the edge of the Quarisen in 1951. The new Neumayer III station, which was completed in February 2009, is located close to the Atka-Bay and is about 16 km south of the

(22)

1.2. Region of interest

shelf ice edge (Roland 2009, p. 47). The Ekströmisen is conned by the ice ridges Søråsen in the west and Halvfarryggen in the east.

Figure 1.1.: Region of interest, hinterland of the German overwintering station Neumayer III. The South African overwintering station Sanae IV is located in the most eastern part of the map. Contour lines are located every 150 m and were derived from a combination of a local elevation model (Drews et al. 2009), an Antarctic-wide elevation model (Bamber et al. 2009) and Landsat photoclinometry data (unpublished data:

NASA's Goddard Space Flight Center). The Moderate Resolution Imaging Spectrora- diometer (MODIS) grounding line (estimate of the region which seperates the grounded ice from the oating ice, see Chapter 1.2 for further explanations) is available to the public and can be downloaded at the National Snow and Ice Data Center (NSIDC) as well as the MODIS Mosaic of Antarctica (MOA) from which the ice divides were evaluated.

(23)

1. Introduction

1.3. Basics of ice dynamics

An ice sheet (Figure 1.2) is dened as a mass of grounded ice which covers a whole continent or sub-continent with an ice thickness high enough to hide most of the underlying bedrock (Bentley and Thomas 2007, p. 100). Like most glaciers an ice sheet is fed by snow accumulation on its surface. As subsequent layers of snow build up, the accumulated snow becomes more compacted rn and is nally transformed into glacier ice. The ice is driven by gravity and ows downhill from the highest points of the interior towards the ocean (Bentley and Thomas 2007, p. 102). In general, three dierent mechanisms of glacial ow can be distinguished (Cuey and Paterson 2010, p. 223): viscous-plastic deformation of ice, sliding of ice over the bed and deformation of the bed itself. Viscous-plastic deformation of ice is characterized by laminar ow. The vertical velocity prole of viscous-plastic ice deformation decreases with depth, but remains constant in the upper parts of the moving ice. Sliding of ice and deformation of the bed are often linked to each other when displacement occurs at the margin between the ice and a deformable bed (Cuey and Paterson 2010, p. 223). If sliding of ice occurs on a rigid bed, there must be a thin layer of water on which the ice starts to slide. This thin layer of water occurs if the pressure on the bottom of the glacier is high enough to shift the melting point below the temperature of the surrounding ice (Meyer 2004, p. 32). Further, it can orginate from geothermal heat ow or a combination of both factors. Deformation of the bed occurs when moving ice deforms a soft sedimentary bed, which limits the forces acting between bed and ice. Since both mechanisms occur at the bottom, the term basal slip can be used both for sliding of ice and deformation of the bed (Cuey and Paterson 2010, p. 223). How these mechanisms of ice motion combine is highly dependent on the thermal properties of the ice and the properties of the underlying bed. Plastic deformation of ice produces relatively slow ice uxes, whereas basal slip mechanisms enable faster ow velocities. The ow of an ice sheet is characterized as slow to moderate, varying gradually with distance... (Cuey and Paterson 2010, p.

356), which suggests viscous-plastic deformation of ice in most places. The velocity of the nonchanneled, slow moving part of the ice sheet covering West Dronning Maud Land is given as 1-15 m/a (Hambrey and Alean 2004, p. 92). However, there are also ice streams, i.e. regions where the ice moves much faster than in the immediate vicinity. In a broader sense, ice streams include fast-owing outlet glaciers (Cuey and Paterson 2010, p. 360). Most ice streams ow along channels in the bedrock with crevassed shear zones separating them from the surrounding slow moving ice (Figure 1.2). The velocity of an ice stream increases towards the coast. Winsborrow et al. (2010, p. 57) suggest that it is not a single factor that governs the ow of an ice stream. In fact, topographic focusing, a soft sedimentary bed, subglacial meltwater and calving margins are believed to be the main inuencing variables for fast ice ow.

In Antarctica, the owing ice sheet reaches the ocean at some point to form ice shelves

(24)

1.3. Basics of ice dynamics

along large parts of the coast. Ice shelves are fed by ice ow from the ice sheet, snow accumulation on the top and basal freezing of the underlying sea water. Ablation of ice shelves is caused by calving of icebergs and basal melting. If an ice shelf is grounded at some point, a dome-shaped ice rise may occur. Typical ow velocities of ice shelves range from a few hundred meters per year up to one kilometer per year at the front. Except for ice rises at a grounding point, there is no ow resistance at the bottom since the ice shelf is oating. Therefore, the ice shelf ow is governed only by a combination of longitudinal and side drags (Cuey and Paterson 2010, p. 373).

Figure 1.2.: Simplied depiction of the Antarctic ice system. Modied after Bell (2009, p. 36).

The transitional zone between the oating ice shelf and the grounded ice sheet is called grounding zone. Figure 1.3 shows a diagram of the grounding zone. The forces driving the motion of ice change drastically here, since the ice begins to oat and basal melting occurs in the grounding zone.

The grounding zone can be dened as the area between F and H in Figure 1.3 and is typically a few kilometers wide (Sykes et al. 2009, p. 35). F indicates the position which limits the ice exure from tidal movement and H indicates the point where the ice starts to oat in hydrostatic equilibrium. The grounding line is the line along the bedrock where the ice starts to oat, indicated as G in Figure 1.3, whereas I is the inexion point where the ice is pressed below the hydrostatic level due to longitudinal stresses.

It is important to monitor the grounding line position, since its location may vary due to changes in ice thickness and sea level (Rabus and Lang 2002, p. 345). Also, ice

(25)

1. Introduction sheet/ocean modeling uses the grounding line as a boundary condition (Sykes et al.

2009, p. 35). Knowledge of the exact grounding zone position is needed for a correct interpretation of interferometric derived velocity data.

Figure 1.3.: Illustration of the grounding zone (after Fricker and Padman (2006, p.

2)).

(26)

(27)

2. Satellite Radar Imaging

Radar systems installed on airplanes or satellites are used for a wide range of appli- cations in earth sciences. A radar system is used as an active remote sensing system which provides its own (microwave) illumination. This gives the radar system the ability to image in daylight or at night. The independence from external illumination is of great interest for the polar regions, since they are covered in darkness for a signicant part of the year. Another advantage of a radar system is that microwaves can penetrate cloud cover. This results in the ability of the radar system to image in nearly all weather conditions. The independence from external illumination and the possibility to penetrate cloud cover are the main advantages of radar systems compared to optical imaging systems.

This section gives a general introduction to radar imaging systems. First, a short de- scription about a radar system with a real aperture (RAR - Real Aperture Radar) is given. Secondly, the advantages of a Synthetic Aperture Radar (SAR) are discussed.

This leads nally to SAR interferometry and its applicability in earth sciences. These principles are used later to derive surface velocities of ice in the hinterland of the Ger- man overwintering station Neumayer III.

2.1. Real Aperture Radar (RAR)

A radar system measures the Two Way Travel Time (TWT) of a microwave pulse.

On airplanes or satellites, nadir¹-looking radars are used as altimeters which measure the distance between platform and ground via the conversion of the TWT to distance.

After the time ∆t the sensor of a radar system receives a part of the energy which was reected from the ground. ∆tis given by

∆t= 2r

c (2.1)

1Nadir: Direction pointing directly below the sensor.

(28)

2.1. Real Aperture Radar (RAR)

where 2r is twice the distance between sensor and ground and c is the speed of light in the atmosphere. In a nadir looking mode, the system does not produce images because of the left-right ambiguity in runtime. In a side-looking geometry, images can be acquired as this ambiguity does not exist. A side looking radar uses a narrow microwave pulse which spreads perpendicular to the ight direction. The wavefront moves over the earth's surface in range direction² and the reected signals are saved as rows in the radar image, relative to their travel time. After that, the radar system moves slightly along the azimuth direction³ and transmits the next microwave pulse.

The reection of this microwave pulse is recorded in the same way as before, but is represented in the next column of the radar image (Albertz 2007, p. 56 f.). Equation (2.2) shows that the resolution in range (Rr_ra) is strongly dependent on the pulse duration t_p of the transmitter

Rrra= c· tp

2 . (2.2)

The longer the pulse, the more targets are hit simultaneously and are thus indistin- guishable. Thus, it is evident that the shorter the duration of the microwave pulse the better the achievable resolution in range. However, due to technical reasons a suciently long pulse duration is needed to achive a reasonable Signal to Noise Ratio (SNR) of the backscattered signal.

The resolution in azimuth direction Rr_az is traditionally dened by the width of the illuminated footprint⁴. The width of the footprint Rr_az is highly dependent on the aperture angle αr and increases with the distance between sensor and ground. The aperture angle α_r is a function of the wavelength λ and antenna length L_r (Meyer 2004, p. 13). The resolution in azimuth direction Rr_az can be approximated by

Rr_az =r· λ

L_r =r·α_r. (2.3)

Equation 2.3 quanties the interrelationship of the resolution in azimuth direction, the distance between satellite and ground and the length of the antenna. If ERS would map with a real aperture and an antenna length of 10 m, it would have an azimuth resolution of about 4 km. This means that only objects separated by at least 4 km in azimuth direction would be resolved. Consequently, the satellite should have a long antenna to achieve an acceptable resolution, which is a dicult technical task. A real aperture radar can be used eectively on airplanes, which do not have that much ground distance when compared to satellites. A synthetic aperture is the solution for a satellite-based system, to get an acceptable resolution in azimuth direction.

2Range direction: Direction perpendicular to the ight direction.

3Azimuth direction: In ight direction.

4Footprint: Ground area which gets illuminated by the satellite's signal.

(29)

2. Satellite Radar Imaging

2.2. Synthetic Aperture Radar (SAR)

As outlined above, a satellite system with a real aperture would achieve a coarse resolution in azimuth direction and is therefore not used on platforms in space. Re- mote sensing satellites like ERS-1 or the space shuttle `Endeavour' during the Shuttle Radar Topography Mission (SRTM) used a SAR.

A SAR system is a coherent radar system which records the amplitude and phase of the backscattered signal. Figure 2.1 describes the relation between phase, amplitude and wavelength. In order to increase the spatial resolution in azimuth direction, several post-processing steps generally referred to as SAR processing, or focusing, are necessary (Massonnet and Feigl 1998, p. 442). SAR processing is a complex procedure, extensively discussed in the literature (see, for example, Cumming and Wong (2005)) and only the basic theory will be discussed here.

Figure 2.1.: Relation between phase, amplitude and wavelength. The amplitude of the backscatterd signal complies with the brightness of the SAR image. The wavelength λis xed for a SAR.

To image in high-resolution along track a long radar antenna is simulated by a SAR.

A relatively small antenna is used to transmit microwave pulses in a wide, club like manner. As the satellite moves between the transmission of two radar pulses, a point on the ground (see point P in Figure 2.2) gets illuminated several times during the pass of the satellite when the distance between the satellite positions of two successive pulse transmissions is much smaller than the length of the footprint (Rr_az). The SAR processing results in an image that can be imagined to be a coherent stack of the overlapping footprints. This means that the Doppler history of the individual targets within the scene is taken into account during the stacking procedure. The length of

(30)

2.2. Synthetic Aperture Radar (SAR)

a synthetic antenna can be described by the dierence between the rst and the last illumination of a ground target. IfS(t₁)in Figure 2.2 is the satellite position where P is illuminated the rst time by the transmitted radar signal and S(t2)the position of the last time, the length of the synthetic apertureL_s can be dened byS(t₂)−S(t₁).

Figure 2.2.: The principle of a SAR system, with S(t1) and S(t2) as the satellite positions where P gets illuminated the rst and the last time. L_srepresents the length of the synthetic aperture.

For a SAR system, the resolution in azimuth direction can be described by (Meyer 2004, p. 14)

Rs_az =r· λ

2L_s. (2.4)

The width of the footprint (Rr_az) is given in equation (2.3). As a result (2.4) can be rewritten as

Rs_az =r· λ 2r· _L^λ

r

= L_r

2 . (2.5)

According to (2.5), the resolution in azimuth direction increases with decreasing an-

(31)

2. Satellite Radar Imaging tenna size of the real aperture. This is at rst counterintuitive, since for a RAR system a longer antenna leads to a better resolution in azimuth direction. However, in light of the coherent stacking procedure it makes sense since a larger footprint (e.g. a smaller antenna) covers the individual targets within the scene more often. In reality, relatively large antennas need to be installed on platforms in space to achieve sucient energy in the backscattered signal (Meyer 2004, p. 14).

The coherent processing of the SAR data produces Single Look Complex (SLC) images. Such a SLC image can be regarded as a projection of the three-dimensional world into an x, y plane, where x is dened by the azimuth direction and y by the range direction. Every pixel in this image has a complex value and contains information of the phase and the amplitude of the backscattered signal.

A SAR image can be distorted by topographical eects. For example, layover eects may occur in mountainous areas: if two targets in a pulse line are within the same distance to the sensor, the resulting values in the SAR image are a mixture of both (Massonnet and Feigl 1998, p. 444). However, these eects can be neglected in the region of interest, since the area is relatively at.

The phase which is saved in an SLC image appears as a value between 0^◦ and 360^◦. For a single SLC image, the phase information is random due to the random scattering from the surface. The phase information of a single SAR image is therefore not of much use and can only be evaluated in a geophysical way when two SAR images are dierentiated. SAR interferometry quanties the dierence between two complex SAR images, which is explained in the next section.

2.3. Interferometric SAR

The previous section discussed how large amounts of raw data can be processed to one high resolution SAR image, via the coherent post-processing of multiple (overlapping) footprints. These SLC images contain information of phase and amplitude for every pixel. The phase dierence of two SLC images can be linked to topography and displacement. This is the main idea of the Interferometric SAR (InSAR) processing and will be discussed in more detail.

For InSAR to work, at least two SAR images of the same area acquired with a slightly dierent view angle need to be available. Figure 2.3 shows the geometrical setup for across-track SAR interferometry, whether for a single-pass⁵ or a repeat-pass⁶ system.

The points i and j are the positions of the SAR during the image acquisition of the same area. The SAR positions i and j are separated through the spatial baseline B_ij. Dierencing the phases of two SLC images acquired with a sensor constellation

5Single-pass system: Both SLC images are acquired at the same time by two SAR sensors.

6Repeat-pass system: The SLC images are acquired at dierent times.

(32)

2.3. Interferometric SAR

as shown in Figure 2.3 results in the interferometric phase ∆φ_ij. The interferometric phase ∆φ_ij is dependent on the dierence in path length⁷ ∆r = r_i − r_j, and can therefore be approximated by

∆φ_ij =φ_j−φ_i = 4π

λ ∆r (2.6)

where φ_i and φ_j are the phase values of the SAR images acquired at the satellite positions i and j. Equation (2.6) can only be valid if the random scatter on the ground is equal for φ_i and φ_j. If this requirement is fullled, which is surprisingly often the case in Antarctica, the random scattering can be removed.

Figure 2.3.: Setup for interferometric imaging, the points i and j are the positions of the SAR during the data acquisition of the same area. θ is the look angle and Bij

the spatial baseline between the two SAR positions.

Equation (2.6) shows, that all contributions which aect ∆r are reected in ∆φij. For a repeat-pass system, as used in this thesis, the single factors contributing to the interferometric phase ∆φ_ij may be approximated by

∆φ_ij =∆φ_orbit+∆φ_topography+∆φ_motion+∆φ_atm+∆φ_noise. (2.7)

7Path length: Distance between SAR and the surface.

(33)

2. Satellite Radar Imaging To get information about surface motion (∆φ_motion) and topography (∆φ_topography) the other phase contents need ideally to be zero.

∆φatm :is the phase dierence due to atmospheric propagation delays. As the state of the atmosphere is not identical between image acquisitions, the interferometric phase is aected by an atmospheric propagation delay. The atmospheric propagation delay is discussed in more detail by Massonnet and Feigl (1998, p. 447).

∆φ_noise : is the phase dierence due to the two random scattering components on the snow covered surface. This component results most likely from unstable surface conditions between the dates of data acqusition (e.g. melting or accumulation of snow, rapid ice movement). A fundamental principle of InSAR to work is that the terms ∆φ_atm and ∆φ_noise need to be minimized. If the terms ∆φ_noise and ∆φ_atm are too large a proper InSAR processing cannot be guaranteed. Therefore, the coherence needs to be calculated to check the suitability of the specic SLC images for the InSAR processing.

The coherence γ is a statistical value which can be described as the degree of decor- relation between two complex SAR images and is expressed by (Rosen et al. 2000, p.

349)

γ = hg1g2∗i

ph|g1|²ih|g2|²i (2.8) where g₁ and g₂ is the backscattered signal received at the respective SAR antenna andh·i denotes the expected value. In practise, the latter is approximated by spatial averaging. The coherence γ is dened for the range between [0, 1], where γ = 1 represents the maximum degree of coherence and γ = 0 the minimum degree of coherence (Meyer 2004, p. 25). As a rule of thumb, a coherence value of 0.3 is noisy but still usable for SAR interferometry, whereas a value of 1 represents excellent coherence, but is very rare (Massom and Lubin 2006, p. 69).

∆φ_orbit :describes the phase dierence due to the dierent acquisition geometry of the SAR sensors. If the acquistion geometry is known, ∆φ_orbit can be simulated and thus be removed from the interferogram with the help of a reference ellipsoid. If this has been done correctly ∆φ_orbit= 0 can be assumed.

After the removal of∆φ_orbitthe interferometric phase∆φ_ij of a coherent interferogram can be approximated by

∆φij =∆φtopography+∆φmotion (2.9)

where the phase dierence induced by topography (∆φ_topography) is dependent on the spatial baseline B_ij and the phase dierence induced by motion (∆φ_motion) is dependent on the temporal baseline∆T.

In Figure 2.3 the spatial baseline B_ij is composed into B_k and B⊥ by projecting position j onr_i. Since

α_ij + (90^◦−θ) + 90^◦+ε= 180^◦, ε=θ−α_ij (2.10)

(34)

2.3. Interferometric SAR

B_k can be dened byB_ijsin(θ−α_ij)and B⊥ byB_ijcos(θ−α_ij). According to Rosen et al. (2000, p. 345) (2.9) can be approximated by

∆φ_ij = 4π

λ B_ijcos(θ₀−α_ij) z

ρ₀sin(θ₀) + 4π

λ ∆ρ (2.11)

where θ₀ describes the look angle to a constant reference surface and ρ₀ indicates the radar range to the reference surface. The topographic height above the reference surface is given by z. Surface displacement between the dates of data acquisition is indicated by∆ρ and will be discussed later.

The sensitivity of the specic SLC image pair to topography depends mainly on the magnitude ofB⊥ and can be approximated by the altitude of ambiguity. The altitude of ambiguity quanties the change in topography needed to induce a phase shift of 2π. The altitude of ambiguity is dependent on the perpendicular baseline B_⊥ and is given by (Massom and Lubin 2006, p. 50)

z2π = λ 2

rsin(θ)

B⊥ (2.12)

where r is the range distance between sensor and target. It is evident from (2.12) that the sensitivity to topography increases with the magnitude of the perpendicular baselineB⊥.

The motion-induced interferometric phase component ∆φ_motion in (2.9) is related to the displacement of the earth's surface between the times of data acquisition. This component can only be measured by a repeat-pass system since a time dierence between image acquisitions at position i and j (Figure 2.3) is required. If this is the case, motion of ice in the satellite's Line Of Sight (LOS) can be detected. A displacement in general is calculated by

δ =x(t_j)−x(t_i) (2.13)

where, for example, an ice particle which is located on positionxat time(t_i)moves to a new position x(t_j)at some later time. To get the average velocity v betweenx(t_i) and x(t_j), the displacement δ is divided by the time period ∆T, which is calculated by ∆T = t_j −t_i. If t_i and t_j in (2.13) are the times of image acquisition at the satellite's position i and j (Figure 2.3), the displacement which is detectable by the satellite can be described by (Kwok and Fahnestock 1996, p. 190)

∆ρ=v·r∆T.ˆ (2.14)

Since the satellite detects only the displacement along its LOS, v is projected onto the LOS byv·rˆwhere rˆis the unit vector pointing from the ground target towards the satellite. As a result, ∆φ_motion in (2.7) is described by

∆φ_motion = 4π

λ ∆ρ. (2.15)

8Vectors are indicated in bold font throughout this thesis.

(35)

2. Satellite Radar Imaging As the look angle θ is relativley steep for ERS (∼23^◦ at the scene center), the sensitivity is greater for vertical displacement than for horizontal displacement (Figure 2.4). According to Figure 2.4 is the sensitivity to horizontal displacement given by

∆ρ= sin(θ)·∆hor and to vertical displacement by ∆ρ= cos(θ)·∆ver.

Figure 2.4.: Sensitivity of ERS to vertical (∆ver) and horizontal (∆hor) motion (after Meyer (2004, p. 29)).

For a 2π phase shift, this leads to

H2π = λ

2 sin(θ) (2.16)

for horizontal motion and to

V2π = λ

2 cos(θ) (2.17)

for vertical motion. As a result, for ERS a motion-induced phase shift of2π is related to horizontal displacement of 7.24 cm or vertical displacement of 3.07 cm at the scene center (Rack et al. 2000, p. 206). This shows the very high sensitivity of this method towards surface displacement.

If there were no spatial baseline B_ij, the phase dierence would represent the ice displacement without any topographical component. Since passing the exact spot twice is technically impossible the phase information∆φ_ij of an interferogram which is constructed with a satellite constellation as shown in Figure 2.3 consists of a topographical part which is dependent on the spatial baselineB_ij and a part representing the displacement in the LOS direction (towards or away from the sensor) which is dependent on the temporal baseline∆T. One has to keep in mind, that small baselines are more sensitive to motion mapping, whereas large baselines favour topography.

(36)

(37)

3. Overview of basic datasets

In this chapter, the datasets on which the derived ow elds are based are introduced. In the rst place, the ERS data are presented, followed by external Digital Elevation Models (DEMs). DEMs are an important input parameter for the interferometric approach which is used to derive surface velocities. This can be understood by looking at (2.9), where a subtraction of ∆φ_topography would lead theoretically to pure displacement along the satellite's LOS. In this study, ∆φ_topography is simulated from dierent datasets. For this, available DEMs in the region of interest are ana- lyzed and compared with laser altimetry data and Global Positioning System (GPS) measurements. Also, GPS-based ow velocities are introduced. These are used for the adjustment and evaluation of the nal ow velocities.

3.1. European Remote Sensing Satellite - 1/2

ERS-1 and ERS-2 are two almost identical earth observation satellites run by the European Space Agency (ESA). ERS-1 was launched in July 1991, followed by ERS-2 in April 1995 (D'Elia and Jutz 1997, p. 1). The most important instrument onboard both satellites is the Active Microwave Instrument (AMI). The AMI is a C-band instrument which combines a SAR and a Wind Scatterometer. The SAR can operate in image and in wave mode. In image mode, a wide swath of about 100 km is recorded, while in wave mode smaller images of about 5 km x 5 km are produced. The wave mode is used to measure length and direction of ocean waves and is of no further concern here, as the data used were solely recorded in image mode. The SAR within the AMI is a side-looking aperture with a look angle of23^◦ (seeθ in Figure 2.3) at the scene center and a wavelengthλ of 5.66 cm. The ground range resolution of the SAR is 20 m across track and 5 m along track. The Wind Scatterometer is used to detect wind speed. Since the AMI is a joint instrument, it is not possible to acquire both types of data at the same time. Another disadvantage is that the SAR instrument requires considerable power and can therefore only be used for 12 minutes per orbit.

The data can only be collected within the range of a suitable ground station since the amount of data is too large for on-board storage (Cracknell 2001, p. 87).

(38)

3.1. European Remote Sensing Satellite - 1/2

Figure 3.1.: ERS-1 & ERS-2 satellite tracks used for InSAR processing. Table 3.1 gives additional information about the shown satellite tracks. In the background is the NSIDC MODIS mosaic of Antarctica (MOA).

ERS-1 and ERS-2 were in a near-polar orbit at a height of ∼780 km until ERS-1 was shut down in March 2000. The satellite data of interest were recorded during the second Ice Phase, lasting from January 1994 to April 1994, and the ERS Tandem Mission which started on 17. August 1995. The rst and second Ice Phases were carried out by ERS-1 only with a repeat pass cycle of three days. Three days is a relatively good time period for interferometric processing, since the surface conditions do not change signicantly and a good coherence between two repeat passes can still be expected. During the ERS Tandem Mission, ERS-1 and ERS-2 orbited the Earth with a time dierence of only one day. In terms of good coherence, this time span is even better than the temporal baseline of the two Ice Phases, however it is less

(39)

3. Overview of basic datasets sensitive to surface displacement. The ERS satellite tracks which were used for the interferometric derivation of ice ow are shown in Figure 3.1.

Table 3.1 gives additional information to Figure 3.1. The track and frame number used, the date of data acquisition and the direction of the satellite pass are listed here.

The latter is important for the generation of a three-dimensional velocity eld, as the satellite only detects motion in its LOS. The combination of an ascending satellite pass and a descending satellite pass gives the user information of two-dimensional ice ow (see Chapter 4.5). The ERS satellite data is available either as ESA processed SLC data or was processed from raw data for the generation of a local InSAR DEM (Drews et al. 2009).

Table 3.1.: ERS-1 & ERS-2 satellite tracks as shown in Figure 3.1. The ocial ESA track and frame number are listed below, as well as the date of data acquisition and the direction of the satellite pass.

ID Track Frame Date Pass

1 493 5121,5103,5085 18/19 Feb 1996 Descending

2 221 5121,5103 05/06 Mar 1996, 09/10 Apr 1996 Descending

3 178 5121,5103 06/07 Apr 1996 Descending

4 035 5085 12/13 Mar 1997 Descending

5 031 5661,5679,5697 06/09 Mar 1994 (2nd Ice Phase) Ascending 6 045 5661,5679,5697 13/14 Mar 1997, 22/23 Feb 1996 Ascending

7 002 5661,5679,5697 15/16 Jan 1996 Ascending

8 188 5697 03/04 Mar 1996 Ascending

9 460 5697,5715 22/23 Mar 1996 Ascending

3.2. Digital Elevation Models (DEMs)

Accurate DEMs play an important role for the generation of three-dimensional ow elds. The Space Shuttle Radar Topography Mission (SRTM) which was carried out in February 2000 acquired elevation data with high spatial resolution. Unfortunately, there are no SRTM datasets for the polar regions as data acquisition took place between 60^◦ latitude north and 56^◦ latitude south only (Massom and Lubin 2006, p.

44). Therefore, other sources need to be employed to get elevation data for the polar regions. An overview of the available DEMs in the region of interest is given in the following.

ASTER GDEM: The Aster GDEM was released in 2009 in a cooperation between the Ministry of Economy, Trade and Industry of Japan (METI) and the National Aeronautics and Space Administration (NASA). The imaging instrument Advanced Spaceborne Thermal Emission and Reection Radiometer (ASTER) uses 14 spectral bands for image acquisition, amongst others a near-infrared band. The near-infrared

(40)

3.2. Digital Elevation Models (DEMs)

band additionally is acquired using a backward-looking telescope. Therefore, along track topographical mapping is possible using a stereo-correlation method. The generated elevation data has a spatial gridding of 30 m x 30 m.

Bamber DEM: The `Antarctic 1 km Digital Elevation Model (DEM) from Combined ERS-1 Radar and ICESat Laser Satellite Altimetry' (Bamber et al. 2009) (hereafter referred to as Bamber DEM) is a combination of laser altimetry measurements from ICESat's Geoscience Laser Altimeter System (GLAS) and satellite radar altimetry data, acquired during the geodetic phase of ERS-1, which started in September, 1994 (D'Elia and Jutz 1997, p. 2). ICESat laser altimetry data has a very good vertical resolution but a poor spatial resolution, while the ERS-1 radar altimeter data has a good spatial coverage, but a poorer vertical resolution. For the Bamber DEM, a spatial gridding of 1 km x 1 km was chosen (Bamber et al. 2009, p. 101). The Bamber DEM is available to the public and can be downloaded at the National Snow and Ice Data Center (NSIDC). At the moment, it is considered to be the most accurate Antarctic-wide elevation model.

Landsat DEM: The Landsat DEM was derived by photoclinometry and has a spatial gridding of 15 m x 15 m. Photoclinometry is a technique which quantitatively relates the brightness of a visible or near-infrared pixel in a satellite image to surface reectivity and local slope orientation with respect to the sun (Massom and Lubin 2006, p. 235). The photoclinometry data was derived from Landsat imagery and GLAS data was used to determine the photoclinometry scaling coecient. The data is available within the Antarctic Surface Accumulation and Ice Discharge (ASAID) (Bindschadler 2007) project at NASA's Goddard Space Flight Center. If the photocli- nometrical approach works, accurate elevation data can be derived with this method.

An example is shown in Figure 3.5. Photoclinometry data is available only for small coastal areas since a lack of contrast is observed in more continental areas. Areas of available photoclinometry data in the region of interest are colored green in Figure 3.3.

Local InSAR DEM: The local InSAR DEM was generated in 2009 at the Alfred- Wegener-Institute for Polar and Marine Research (AWI). It has a spatial gridding of 50 m x 50 m and covers most parts of the survey area, except for the oating shelf ice.

It is based on an InSAR approach using SAR data from ESA's ERS-1/2 including the SAR images shown in Figure 3.1. The interferometric derived elevation information was combined with laser altimeter data from ICESat's GLAS (Drews et al. 2009).

RAMP DEM: The Antarctic-wide Radarsat Antarctic Mapping Project (RAMP) DEM is a combination of many dierent methods, amongst others: GPS, satellite ERS-1 radar altimetry and Radio Echo Sounding (RES). The spatial gridding is given as 200 m x 200 m, but parts of the model have a spatial gridding of 5 km x 5

(41)

3. Overview of basic datasets km due to the dierent input datasets. The RAMP DEM is available to the public and can be downloaded at the NSIDC.

Wesche DEM: The local Wesche DEM was released by C. Wesche in 2009 (Wesche 2009). The region of interest is completly covered by this DEM. The spatial gridding is 2.5 km x 2.5 km. The DEM is interpolated through kriging and orginates from dif- ferent data sources, namley: kinematic GPS measurement, airborne radar altimetry, RES and GLAS laser altimetry.

3.3. Ground control and validation

The general inaccessibility of the Antarctic continent hampers the generation of high quality DEMs, which are by now standard for other parts of the world. For Antarc- tica, dierent remote sensing techniques are applied, each having its pros and cons, depending on the specic terrain. This is a challenge for every Antarctic-wide DEM which relies on one method only. Altimetry works great on the at Antarctic plateau, but runs into problems in areas with higher surface slopes. Photoclinometry fails if the albedo changes due to varying snow conditions. Interferometry-derived DEMs are challenged by uncertain satellite trajectories and atmospheric conditions, and the list continues. Therefore, it is crucial to evaluate the DEMs with Ground Control Points (GCPs), if they are available. This section introduces the datasets which were used for an evaluation of the DEMs and for validation and adjustment of the nal derived velocity elds. Airborne laser altimetry data is introduced rst, followed by kinematic GPS measurements. Both datasets were used for the evaluation of the DEMs used. Kinematic GPS measurements were used for the validation and calibration of the calculated ow velocities.

Airborne laser altimetry data: The laser altimetry data of interest were recorded with an Airborne Laser Scanner (ALS) which was installed on the scientic aircraft Polar 5 in 2007. The ALS was operated with 80 Hz and a scan angle of 45^◦ (Helm et al. 2007, p. 2). The footprint of the ALS was about 1 m along track and 6 m across track. The deviation of the laser altimetry data to GPS measurement is within the range of centimeters. The ALS elevation data which is used in this study was interpolated to a 50 m x 50 m grid and serves as a precise reference base for the evaluation of the DEMs.

Kinematic GPS measurements: The kinematic GPS dataset shown as prole 2 in Figure 3.3 was recorded during a eld campaign in January and February 2007. Local reference stations were used for the data processing (Wesche et al. 2009, p. 382). The

(42)

3.3. Ground control and validation

distance between the data points is given with 3 m. The vertical accuracy in elevation of the kinematic GPS data is in the order of centimeters to meters depending on the length of the baseline to the reference station.

As the region of interest is completely covered with ice, it is hard to nd exposed bedrock where the surface displacement can be assumed to be zero. Such tie-points would be very valuable for the calibration and control of the satellite-derived ow velocities. Fortunately, sporadic GPS-derived velocity measurements are available in the region of interest (Riedel 2002, p. 66). The selected GPS velocity measurements are shown as arrows in Figure 3.3 and are listed in Table 3.2. The GCP HALVFAR in Table 3.2 was acquired in the same eld campaign as the kinematic GPS measurement used for the DEM comparison (pers. comm. C. Wesche). The other GCPs listed in Table 3.2 were acquired during the Polarstern cruise ANT XIV/3. In this scientic cruise, an geophysical-geodetic eld survey took place at the grounding zone of the Ekströmisen ice shelf (Riedel 2002, p. 66). One geodetic goal of the survey was to measure the response of the ice body to the ocean tides at dierent locations.

Kinematic GPS measurements were carried out in the grounding zone in connection with a reference station on solid rock (Riedel et al. 1999, p. 240). Three-dimensional movement of the antenna positions could be derived from these measurements (Riedel 2002, p. 67). The velocity of the ice ow was measured as 27 m/a 30 km south of the grounding zone and varied between 75-148 m/a at the grounding zone. On the oating ice shelf, the ow accelerates up to 222 m/a at the shelf ice edge.

The GPS points GLSS and HALVFAR (red arrows in Figure 3.3) were used for the calibration of the velocity of the grounded ice as the interferogram is most probably not aected by tidal movement in these areas (Chapter 4.2). GPS point 905 (Table 3.2) was used for the calibration of the velocity of the oating ice shelf.

Table 3.2.: GPS-derived ow vectors in the Ekströmisen ice shelf area. Modied after Riedel (2002, p. 66).

Station Latitude [^◦] Longitude [^◦] Flow vector v [m/d] α[^◦] Time of measurement [d]

153 -70.698056 -9.222222 0.639726 325.3 11

305 -70.871111 -8.468889 0.506027 334.0 29

505 -71.044444 -8.477222 0.457534 337.4 13

705 -71.221667 -8.412778 0.458904 351.6 11

905 -71.401667 -8.347222 0.390137 358.5 10

1105 -71.578611 -8.325278 0.398630 22.3 13

1305 -71.720000 -8.521389 0.324384 23.6 12

GLN -71.625278 -8.497222 0.405479 25.7 11

GLNM -71.668611 -8.562500 0.396986 28.5 10

GL0 -71.709167 -8.622778 0.393699 31.4 10

GLSM -71.752222 -8.664444 0.342466 27.0 10

GLS -71.795556 -8.707222 0.205479 21.2 10

GLSS -71.989444 -8.722222 0.075616 3.1 7

HALVFAR -70.92568 -7.391785 0.01254 301.5 46

(43)

3. Overview of basic datasets

3.4. Comparison of DEMs used

In this section, the available DEMs in the region of interest are compared with airborne laser altimetry data. The DEMs which are nally used for simulating

∆φ_topography are additionally compared with ground-based GPS measurements.

Basic error estimation can be obtained by calculating the Root Mean Square Error (RMSE). To calculate the RMSE, a reference dataset needs to be available. In this study, the airborne laser altimetry dataset serves as a reference. For comparison, all datasets were resampled on a 125 m x 125 m grid. 11705 points derived from the laser altimetry dataset were used to calculate the RMSE for the available DEMs (Table 3.3). This value represents n in equation (3.1).

RM SE =

r Pn

n=1(i1−i2)²

n . (3.1)

In (3.1), i₁ is the same point from the DEM as i₂ from the laser altimetry dataset.

Table 3.3 shows the RMSE for the available DEMs, except for the Landsat photoclinometry data, since the ALS data does not cover the Landsat DEM suciently.

Table 3.3.: Available DEMs for the region of interest. The spatial gridding is shown together with the RMSE based on airborne laser altimetry data. Also the coverage, method and source of the DEMs are listed.

Name GRID RMSE Coverage Method Source

ASTER GDEM 30 m 894.9 m World-wide Stereo-correlation NASA

Bamber DEM 1 km 40.5 m Antarctic-wide Laser-, radar altimet. NSIDC

Landsat DEM 20 m - Coastal areas Photoclinometry NASA (unpubl.)

local InSAR DEM 50 m 12.3 m Local SAR interferometry AWI

RAMP DEM 200 m 177.3 m Antarctic-wide GPS, radar altimet., RES NSIDC

Wesche DEM 2.5 km 24 m DML GPS, laser-, radar altimet., RES AWI

The RMSE shown in Table 3.3 varies strongly for the available DEMs. This shows the importance of testing the DEMs beforehand, if possible. For example, the RMSE of the global ASTER DEM is given as 21.19 m in the ASTER Global DEM Validation Summary Report (ASTER GDEM Validation Team 2009, p. 7). However, the RMSE in the survey area is about 42 times higher, perhaps due to the bad texture of the snow-covered surface in the survey area.

The calculated ow velocities in this study are based on the Bamber DEM and the local InSAR DEM. This is because of the high quality of both DEMs on the one hand and because of time constraints on the other. Nevertheless, the script which was written to automate the processing of ow velocity elds is designed for an easy

(44)

3.4. Comparison of DEMs used

input of dierent elevation datasets. For example, there are high expectations on the TanDEM-X¹ mission which was started in June 2010 (DLR 2010).

In the following, both elevation models used are compared again with ALS data and additionally with GPS eld measurements to point out the spatial inaccuracy/accuracy of both elevation models in the survey area. The inuence of the external elevation data on the nal ow velocities is shown in Chapter 5.

Figure 3.2 illustrates the dierences of the two DEMs with the available laser altimetry data. Obviously the local InSAR DEM leads to better results in this region, which can already be seen by looking at the margins of the color bars (Figure 3.2, top).

Furthermore, two proles were taken for comparison in two relevant regions (Figure 3.3). Prole 1 is roughly perpendicular to the main ice ow into the Ekströmisen.

Prole 2 is in the summit region of the ice ridge Halvfarryggen, a potential drill site for a deep ice core.

In prole 1, data from the laser altimetry dataset serves as a reference base and prole 2 is based on the kinematic GPS measurements. Figure 3.4 shows a prole of both DEMs along the laser altimetry prole 2 (top) as well as the dierences to the laser altimetry data (bottom). The largest deviation of the local InSAR DEM is -20 m at one point (Figure 3.4, bottom), which again emphasizes the high accuracy of the InSAR DEM. Prole 2 (Figure 3.5) shows the photoclinometry data which was derived from Landsat imagery in comparision to the local InSAR DEM and the Bamber DEM. The photoclinometry data looks promising and appears to be very accurate in places, but is only available for small coastal areas since a lack of contrast is observed in more continental areas and is therefore of no further concern here. Nevertheless, a combination with elevation data from other sources could lead to interesting results (e.g. contour lines in Figure 1.1). The two other DEMs, the local InSAR DEM and the Bamber DEM, come o more badly in this area. This might be due to high accumulation rates in this region in the case of InSAR. High accumulation rates can lead to bad coherence as the surface conditions change signicantly between the dates of data acquisition. For altimetry, this might be due to the relatively high surface slope.

In conclusion, the local InSAR DEM compares better to GCPs than the Bamber DEM, which is nevertheless a very accurate elevation model on a larger scale.

1TanDEM-X: Bistatic SAR mission, with two almost identical satellites ying in a close across-track formation for topographical mapping.

(45)

Figure 3.2.: Dierence in meters along airborne laser altimetry proles. Top left:

local InSAR DEM (Drews et al. 2009), MODIS grounding line in the background. Top right: Antarctic-wide elevation model (Bamber et al. 2009), MODIS grounding line in the background. The deviation is much higher with the Antarctic-wide elevation model on the right-hand side. Bottom: Histograms of the deviation to the laser altimetry data.

(46)

3.4. Comparison of DEMs used

Figure 3.3.: Location of proles for DEM comparison. 1 shows the laser altimetry prole (Figure 3.4) and 2 the GPS-derived prole (Figure 3.5). The arrows represent GPS-derived velocity measurements, and the red arrows indicate the GCPs which were used for velocity adjustment (Chapter 4.4). The green areas denote the availability of photoclinometry data in this region. In the background is the NSIDC MODIS mosaic of Antarctica (MOA).

(47)

Figure 3.4.: Laser altimetry prole from prole 1 in Figure 3.3. Top: Prole along the Bamber DEM, the InSAR DEM and the airborne laser altimetry data. Bottom:

Dierences between the InSAR DEM and the Bamber DEM to laser altimetry data along the prole.

Figure 3.5.: GPS-measured prole from prole 2 in Figure 3.3. Top: Prole along the Bamber DEM, the InSAR DEM, the Landsat DEM and the kinematic GPS measurement. Bottom: Dierences between the InSAR DEM, the Bamber DEM and the Landsat DEM to kinematic GPS measurement along the prole.

(48)

Surface velocities in the hinterland of the Neumayer III station (Antarctica)

Surface velocities in the hinterland of the Neumayer III station (Antarctica)

derived from SAR-Interferometry

Niklas Neckel

Universität Heidelberg

Fakultät für Chemie und Geowissenschaften

Diplomarbeit/Diploma Thesis

Surface velocities in the hinterland of the Neumayer III station (Antarctica)

derived from SAR-Interferometry

December 21, 2010

Abstract

Zusammenfassung

Contents

List of Figures

List of Tables

List of Abbreviations

1. Introduction

1.1. Goals of this study

1.2. Region of interest

1.3. Basics of ice dynamics

2. Satellite Radar Imaging

2.1. Real Aperture Radar (RAR)

2.2. Synthetic Aperture Radar (SAR)

2.3. Interferometric SAR

3. Overview of basic datasets

3.1. European Remote Sensing Satellite - 1/2

3.2. Digital Elevation Models (DEMs)

3.3. Ground control and validation

3.4. Comparison of DEMs used